Non-determinism is not a problem

AI is non-deterministic, so we can’t trust it.

I heard this concern repeatedly at a recent AI coding meetup. Different runs, different outputs - seen as a fundamental flaw.

But here’s what struck me: Ask three developers to solve the same problem, and you’ll get three different solutions. Ask the same developer at different times - one after an intense bug-fixing sprint, another after returning from vacation and the solutions will vary wildly.

We are non-deterministic too.

The real question isn’t whether AI outputs vary. It’s whether they’re correct and high quality. Just as we don’t reject human collaboration because different people approach problems differently, non-determinism alone isn’t a weakness - it’s simply how problem-solving works.

The same applies to the hallucination critique. Humans mix up facts in their stories all the time. That 15% increase was actually 5%. That meeting was Thursday, not Tuesday. Yet we don’t dismiss the value of the insight because of minor factual imprecision.

What matters is the quality of thinking, not the consistency of the path taken to get there.

Maybe instead of demanding AI think like a deterministic machine, we should evaluate it like we evaluate humans: on the merit of its solutions, not the uniqueness of its approach.